Adapt Progress Evolve (@ape)

In depth: Adapt Progress Evolve: A London Software Studio Run by a Fleet of AI Agents

Written from Adapt Progress Evolve's own profile data — claims, answers and testimonials — in a measured human voice via ScriptGrain · last written 2026-10-02

Most software studios have a bench of people, a backlog and a client list. Adapt Progress Evolve has a fleet. Picture a pit lane where the crew is made up of specialised Claude agents, each with one job, and where a human stands at the end of the lane with a hand up, deciding what actually leaves the garage.

That, basically, is the whole idea. The rest is detail, and the detail is worth sitting with.

What It Is

Adapt Progress Evolve is a UK software studio, based in London, that designs, builds and operates its own portfolio of digital products. It's operated by Adapt Progress Evolve Limited, registered in England and Wales (company number 10047328, VAT GB362654684). Jack Stovell is the founder.

Is it an agency? Not primarily. The studio builds and runs its own products, and alongside them takes on selective client work. The order matters here: products first, services second. Most agencies work the other way round.

How the Work Gets Done

Day-to-day work is handled by a fleet of specialised Claude agents. The fleet covers engineering, SEO, security, content, design review and research; each agent has a lane, and the lanes are drawn by function rather than by project.

That said, nothing ships on an agent's say-so. Human approval gates sit on everything that goes out, so the agents do the volume and a person does the signing off. It's a division of labour that treats speed and judgement as separate jobs, which they are.

Its services page is short too: AI agent development, Claude and LLM API integration, and rapid MVP development. Four items, and each one maps onto something the fleet is already doing to its own products.

The Portfolio

The list of live products is longer than you'd expect from a studio this size. ScriptGrain handles writing in your own voice. Launch Video Maker turns websites into short videos. Visibility to ai produces AI answer visibility reports, How to get AI to publishes practical AI guides, and Hilyt builds AI-readable profiles. What is GEO is a free guide to generative engine optimisation. Two more, ukspend (London council spending) and Paws of London (dog-friendly venues), point at local data rather than software tooling.

Forematter is in development. MarketFrame and ReviewNudge are paused while they're rebuilt, which is an honest status to publish; plenty of studios would simply let a stalled product drift quietly off the page.

Two launches from late September 2026 show the pattern in practice.

Launch Video Maker went live on 28 September 2026. It takes a business website and turns it into a short motion reel, with music and sound effects, priced from £9 per video. A specific input, a specific output at a specific price.

Assets for three.js sells original GLB model packs built in Blender for use with three.js. As of 30 September 2026 the catalogue stood at 312 assets across 10 packs, sold once per pack rather than by subscription. So the customer pays, downloads and owns the thing.

The Open-Source Side

Not everything the studio makes is sold. Parts of its stack are open-sourced, including Cortex Map, a Three.js knowledge-graph component for React. Note the overlap: Three.js turns up in the open-source work and in the asset packs, which suggests the studio is building depth in one corner of the web rather than scattering itself across all of it.

Who It's For

The audience splits in a few directions, and it helps to say so plainly. Small businesses and creators are the customers of the individual products: someone who needs a promo reel from their website, or writing that sounds like them, or a set of 3D models for a project. Developers are the audience for the open-source tooling and the asset packs. And organisations wanting something built, or wanting an agent fleet set up, are the audience for the selective client work.

Each group is served by a different door, but the same operating model sits behind all of them.

What to Make of It

The interesting question isn't whether AI agents can write code; that argument is largely settled. It's whether a small studio can run a broad portfolio (nine live products, one in development, two under rebuild, plus open source) without the usual overhead of a large team. Adapt Progress Evolve is testing that in public, with its own products as the evidence.

The agents supply the throughput. The approval gates supply the accountability. And the fleet, like any good pit crew, only earns its keep when a human is still deciding when the car goes back out.